Latest / AI For Humans: Weekly AI News, Tools & Trends / OpenAI Rips YouTube, Elon's Robotaxis & AI Artist Purz Beats | Ep52
Transcript
- 0:00OpenAI is screwing YouTube and
- 0:02It kind of seems like everybody is screwing YouTube
- 0:05There's only a few of these that exist in the world, we actually have an AI DJ
- 0:09Let's crank it up and let the beat drop like it's hot, hot, hot.
- 0:15every Experience that we have is gonna be synthesized for a new large language model
- 0:21Now I wish I could punch you through the screen again
- 0:23Just kidding, everybody.
- 0:23We are having fun.
- 0:27Welcome.
- 0:28Welcome.
- 0:28Welcome everybody.
- 0:29It is another big episode of AI for humans.
- 0:32Oh,
- 0:32There was no, there was no warning whatsoever about
- 0:36We'll start over.
- 0:37We'll start
- 0:38And I felt unwelcomed,
- 0:41We're starting over.
- 0:42Stop.
- 0:43We're starting over.
- 0:44Welcome everybody.
- 0:45It is AI for Humans, your weekly guide to the wonderful and
- 0:48wild world of generative AI.
- 0:49I am here.
- 0:50My name is Gavin Purcell and my friend, Kevin Pereira is on the
- 0:53other end of the microphone.
- 0:54Kevin, how are you?
- 0:55I'm on the other end of this microphone.
- 0:57It's like two tin cans in a
- 0:59We're connected.
- 1:00We're connected.
- 1:01That's it.
- 1:02We've docked.
- 1:03Hi friends.
- 1:04me, KP.
- 1:05I'm leaving that entire intro in Gavin.
- 1:07What a beautiful AI for humans.
- 1:09Episode five, two.
- 1:11We got today.
- 1:12Should we tell the people what this podcast is about?
- 1:15We like to demystify all the news, tools, and all the other
- 1:18aspects out there just for you.
- 1:20Kevin, what's on the show today?
- 1:21OpenAI is screwing YouTube and Google is screwing YouTube.
- 1:26It kind of seems like everybody is screwing YouTube, except
- 1:29for maybe the creators.
- 1:30We're gonna tell you how and why, and maybe, maybe I'll share some very
- 1:34artistic, tasteful renderings that I've made of said screwing, Gavin.
- 1:38no, really?
- 1:40No!
- 1:41Is it okay?
- 1:41Good, good.
- 1:42I was just making sure you're paying attention.
- 1:44Also, hey, stop me if you've heard this one, Gavin, and or audience
- 1:48who cannot actually stop me because this is a one way medium.
- 1:52Elon Musk is working on robo taxis.
- 1:56Oh, I've heard this one, so you can
- 1:57Yeah, we've all heard this one since like 2014, but hopefully, the just
- 2:02announced Spotify AI DJ will be able to generate a playlist that's loud enough to
- 2:06drown out the screams as our robo taxis autopilot us through farmer's markets.
- 2:11Oh no, sir, I don't like that at all.
- 2:14. Listen, we're going to have all the details, plus some dumb things that
- 2:17everybody listening can do with AI today.
- 2:20We're going to chat with an AI powered guest, and then we're probably
- 2:23going to have a way more compelling chat with an actual human being,
- 2:26Gavin, a boundary pushing visual artist, who is our real guest today.
- 2:31We're going to dive deep into the cutting edge of AI artistry with PURZ.
- 2:37PURZ.
- 2:38It's PURZ.
- 2:38P U R Z,
- 2:39It's not Perz Beats, I thought his name was
- 2:41that's the full name, but colloquially, the casual, we're on that level.
- 2:45When we give a head nod, we say, Sup, Purrs?
- 2:47what's up Perz?
- 2:48All right, that'll be very exciting.
- 2:49Before we get started as always we want to tell you at home Please
- 2:54share , and rate our podcast.
- 2:56It is only with those shares and rates that we grow.
- 3:00Our YouTube video last week was doing very well and still is doing very well.
- 3:03And we appreciate everybody that watches the video likes and subscribes on YouTube.
- 3:07Also, please leave a five star podcast on Apple podcasts.
- 3:10We will read them at the end of the show.
- 3:12Kevin, today we have three new ones to read, which is exciting.
- 3:14Um, yes, we, we had three new.
- 3:17last week's
- 3:18All right, groveling?
- 3:18through to people?
- 3:19It got through.
- 3:20So again, we really appreciate everybody in our audience who listens to this show.
- 3:24This is not just the two of us BS ing for an hour and then we upload it.
- 3:28There's a lot of editing and all sorts of other
- 3:30hate each other.
- 3:31Every sentence is a grind.
- 3:34It is trench warfare when we launch this pod.
- 3:37And through gritted teeth and multiple takes, we managed to get sentences out.
- 3:41So if you appreciate the end product, we appreciate you engaging.
- 3:45Isn't that right, you piece of shit?
- 3:48Wow!
- 3:49Now I wish I could punch you through the screen again.
- 3:51Just kidding, everybody.
- 3:52We are having fun.
- 3:53I want everybody to know that is not exactly true.
- 3:56Kevin is making up lies right now.
- 3:57Kevin, it is time to get to the news!
- 4:03I'm already sweaty.
- 4:15Okay.
- 4:15The news this week is as usual, Fast and Furious, the biggest story that
- 4:19I've seen come down news story in a while and AI broke over the weekend.
- 4:24The New York times published a piece that had five authors on it.
- 4:27So, you know, when there's five authors on New York times piece, they've
- 4:30done some research, which basically accused open AI, Google, and meta.
- 4:34all of training on, um, various versions of data, but mostly they
- 4:39talked about scraping YouTube.
- 4:40And the big question with Sora, videos especially, has been, how
- 4:44did they get video to train Sora?
- 4:46Because as everybody knows who listens to the show, and if you
- 4:48don't, it's a pretty simple thing.
- 4:50You need a lot of data to train an AI model, whether
- 4:53that's an LLM and text model.
- 4:55It means you need to get a lot of data of words.
- 4:57In this instance, you need a lot of video to train something like
- 5:00SORA, especially train it as well.
- 5:01Now, OpenAI has done a deal with Shutterstock.
- 5:04So there's a lot of stuff out there, but most people, I think,
- 5:07haven't assumed that there was some sort of version of this happening.
- 5:10I think there was a little bit of a snippet that came out a couple weeks ago
- 5:13when Joanne Stern published her video where she interviewed Mina Murati, who
- 5:17is the OpenAI Chief Technical Officer, who could not answer the question about
- 5:21how they had trained their SORA model.
- 5:24She had a little bit of, yeah, that was a little bit of a face.
- 5:28So she really didn't have a good answer for this.
- 5:30And I think what we've learned now is that they did train it on YouTube.
- 5:33So, Kevin.
- 5:34First of all, when you read this piece, which was very long, and I encourage
- 5:37everybody who's listening to go read it.
- 5:39Cause it's a lot of, uh, there's a lot of things in there
- 5:41that are really fascinating.
- 5:42What was your first takeaway from reading this piece?
- 5:48What a non surprise!
- 5:52I wasn't shocked in the slightest.
- 5:54I got to imagine you weren't as well, Gavin, because we know how much data
- 5:58is required to train these things.
- 6:00, a couple of details from the article, and then we can dive in a little bit deeper.
- 6:03Greg Brockman, OpenAI's president, , allegedly personally helped
- 6:07collect the videos that were used.
- 6:09, again, according to the New York Times.
- 6:10What they did though, was that they grabbed, , transcripts
- 6:13of all the YouTube videos.
- 6:14And using that, which would be a violation, according to Google,
- 6:19of YouTube's usage policies.
- 6:20You cannot use their data to train something else.
- 6:24However, buried within that article is an accusation that Google may have done the
- 6:29Done it themselves.
- 6:30Yeah, exactly.
- 6:31and that the reason Google might not be making any public statements
- 6:35about OpenAI's alleged actions is that they're curious if they
- 6:39would be outing themselves as well.
- 6:41Exactly.
- 6:41And it's been really interesting.
- 6:42So Neil Mohan, the, , YouTube CEO did come out and say that, , some YouTube
- 6:48content is scrapable for open web purposes, but the video transcript and
- 6:51footage are not allowed to be scraped.
- 6:52He said, this is a clear violation of our terms of service.
- 6:55So those are the rules of the road in terms of content on our platform.
- 6:58But you're right.
- 6:59It is interesting to see.
- 7:01how how kind of quiet the overall Google ecosystem has been on this.
- 7:06And I think one thing I want to point out
- 7:08I was really kind of surprised at how little noise this
- 7:11made in the AI ecosystem.
- 7:13And when I say the AI ecosystem, I mean, the people kind of like us or more of
- 7:17the people who are kind of interested in AI who are kind of focused on this.
- 7:20When I first read this, I was like, well, this is a smoking gun.
- 7:23Like, look at this.
- 7:24This is going to like launch what I believe will probably be dozens of
- 7:28lawsuits, whether they get through or not.
- 7:30Now, I think the big question is going to be what do Google's terms of services
- 7:35say when things were scraped, if they were scraped, and it sounds like they were.
- 7:38But then I think the next question is going to be.
- 7:40And this, I want to ask you directly this, cause there's so many people
- 7:43who talk about, and you've said this and I've said this, the idea that
- 7:46like the cat is out of the bag, right?
- 7:49That the idea that once you've done this, guess what?
- 7:52It's too late.
- 7:53All this stuff is out there and now we just have to live with what it is.
- 7:56Do you think that there is any way that there is a pause that can be put on
- 8:01this based on legal terms at this point?
- 8:04Perhaps an injunction against the output of all large language models
- 8:08until a judge can say hey We got to piece this together, right?
- 8:11We got to dive in through your model, see how it was trained,
- 8:13see if you violated something.
- 8:15, I think there's too much money at play , that, that could be a scenario, but I
- 8:19think there will just be millions, if not billions of dollars thrown against the
- 8:24wall to make sure that something like that doesn't happen in the interim and it's
- 8:27business as usual until decades later, everything megacorp that's going to rule
- 8:34us all and be our one world government.
- 8:36Like, people will just play nice.
- 8:39There's just too much money at stake, so I think this is a calculated move.
- 8:43I'm sure there were lawyers involved at one point in the room, and
- 8:46eventually, probably the engineers said, Hey, listen, we have to sprint.
- 8:50We have to grab this data.
- 8:52We will beg for, and probably pay for, forgiveness later.
- 8:55So let's just go.
- 8:57I think that's probably true.
- 8:58And I think the one thing that's important to point out here is there's been a couple
- 9:01stories this week about how hard it is to get new data to train , these devices on.
- 9:05And there's lots of people in the machine learning world that are saying, Hey,
- 9:08this is going to be a scaling issue.
- 9:10Like actually every time you scale up, you get better results, but
- 9:13scaling equals more data, right?
- 9:15And so the one thing that they've talked about a lot is we have
- 9:19scraped so much text data already.
- 9:21Like so many books and again, who knows the legal ramifications of that, putting
- 9:26it aside, but we've scraped the internet.
- 9:27We've scraped books.
- 9:28We've scraped all this stuff.
- 9:30If they're scraping and have scraped, , a lot of YouTube already, where does
- 9:34more data come from is a huge question.
- 9:37And that's partly the whole synthetic data conversation.
- 9:39But like, I don't know where you get more than YouTube when it comes to video.
- 9:43Someone has to be transcribing and scraping every podcast in every
- 9:48language and going back , decades ago.
- 9:51, I'm sure that's already been thought of and is being grabbed as well.
- 9:54, last year, this is according to the New York times article, Google quote,
- 9:57Also brought in its terms of service.
- 9:59One motivation for the change, according to members of the company's
- 10:02privacy team and an internal message reviewed by the times was to allow
- 10:07Google to be able to tap publicly available, Google docs, restaurant
- 10:11reviews on Google maps and other online material for more of its AI products.
- 10:15So even Google was like, Hey, we got to start getting this, you know, at
- 10:19some point there's going to be some sort of like, Hey, we'll give you.
- 10:22Maybe, maybe we'll give you 3 percent off your Google for business account.
- 10:26If you let us crawl through some files, in fact, maybe we
- 10:30won't give you anything off.
- 10:32Maybe we'll just
- 10:33We'll just do it.
- 10:34of service and just go on through it.
- 10:36That's probably going to happen.
- 10:37But on the synthetic data front, that, that you brought up, that to me is
- 10:41really, really interesting because.
- 10:43It seems like every day a new paper comes out that says synthetic data
- 10:47is great and it's going to solve the problem, which is something that Sam
- 10:50Altman has said is going to be the case.
- 10:53Or there's a paper that comes out and says synthetic data is terrible and the
- 10:56AI is going to train itself in a loop.
- 10:57So for the broader audience out there that doesn't know what we are talking
- 11:00about when we say synthetic data, we're saying that You can use the large
- 11:04language models as they exist today and likely as they'll exist in the
- 11:07next year or two, which will be better, theoretically, to generate paragraphs upon
- 11:13paragraphs of text and a computer code.
- 11:17I mean, we're talking like generate the Library of Congress.
- 11:19It could probably do it in a week.
- 11:20It just is going to churn and give you fake questions and answers, fake recipes,
- 11:26fake restaurant reviews, fake everything.
- 11:28That synthetic data is thought by some to be good enough.
- 11:33That if you generate enough of it, you can train a better model.
- 11:36But some are saying that actually the repeated words and phrases that are
- 11:41already seen to come out of some of these language models, the bad code
- 11:45or the potentially pirated snippets of content, those are just going
- 11:48to resurface over and over again.
- 11:50And then eventually it kind of trains itself on its worst bad habits
- 11:53and it's not going to be usable.
- 11:55Those are the two schools of thought right now.
- 11:57Do you have a horsey in this race, Gavin?
- 12:00I don't understand enough on the technical side to see if the
- 12:02synthetic data is strong enough.
- 12:03I will say from what I've read, the arguments around synthetic data is it
- 12:07wasn't good enough before, but now as the models get stronger, it may be better.
- 12:12The interesting thing to think about is, if that is the unlock and we're also
- 12:16heading to a world where only bigger, bigger models equal better results,
- 12:20then we are going to be moving very quickly to the next stage of whatever
- 12:24this AI world is, because the AI, as we talked about in the NVIDIA episode,
- 12:28the AI can simulate themselves in different places and different things.
- 12:33And if that allows, if you can simulate.
- 12:36Interactions, text, video, all that stuff.
- 12:39You can ostensibly do that in an infinite level if you have the amount
- 12:43of compute and storage to do it.
- 12:44And then it goes really quickly.
- 12:46Actually, this really dovetails really interestingly into another story that we
- 12:50have Bill Peebles, one of the main open AI engineers behind Sora had a really
- 12:55interesting talk that he came out and gave to the, AGI house, which is again,
- 12:59like probably some sort of hype house, but for AGI is like, whoop, whoop.
- 13:02We're going to AGI land, baby.
- 13:05Like and
- 13:06Energy Refrigerator!
- 13:07It's so hype!
- 13:09Yeah.
- 13:09So Bill was at the AGI house and gave a speech, and we're going to hear a little
- 13:12snippet of what his speech was here.
- 13:15So, of course everyone's very bullish on the role that LLMs are
- 13:18going to play in getting to AGI.
- 13:20But we believe that video models are on the critical path to it.
- 13:24And concretely, we believe that when we look at very complex scenes that Sora
- 13:28can generate, like that snowy scene in Tokyo that we saw in the very beginning,
- 13:33that Sora is already beginning to show a detailed understanding of how humans
- 13:36interact with one another, how they have physical contact with one another.
- 13:40And as we continue to scale this paradigm, we think eventually it's going
- 13:43to have to model a human state, right?
- 13:45The only way you can generate truly realistic video, with truly
- 13:48realistic sequences of actions.
- 13:49is if you have an internal model of how all objects,
- 13:52humans, etc., environments work.
- 13:54And so we think this is how Sora is going to contribute to HDI.
- 13:58Basically what Mr.
- 13:59Peebles is saying here, call him, you know, Mr.
- 14:01Peebles if you're nasty, uh, He's talking about, he's talking about
- 14:06the idea that Sora, which is a video generator, is actually a world
- 14:11simulator, ostensibly, and we've talked about this on the show before,
- 14:14but by world simulation, and by doing that, even in synthetic environments,
- 14:19you are teaching the AI How people and objects interact with each other.
- 14:24And that is a lot of people see as the next stage of data, right?
- 14:27It's like real world training or simulated world training
- 14:30outside of just words and texts.
- 14:33You're talking now about actions.
- 14:34You're talking about the way a tree blows in the wind.
- 14:36You're talking about the way a human interacts with that tree when it blows
- 14:40against them, where they cut it down.
- 14:41All of that stuff is data, right?
- 14:44It can't be perceived as data.
- 14:46And if they can find a way that Sora can generate synthetic models
- 14:49of the world, that does feel like where we're headed next.
- 14:53I think 2026 is the year that And it's going to need a sexier term
- 14:58to probably get the youth involved, Gavin, but, uh, let's say, let's
- 15:01say data harvesters or, aggregators.
- 15:05Basically, we are going to, for pennies per minute, of experience,
- 15:12we are all going to end up working for open AI slash Microsoft.
- 15:15We're going to have our glasses on that are going to feed constant video streams.
- 15:20Maybe we'll have tactile haptic gloves so that it can know the pressure and force
- 15:24with which we're exerting upon the world as we navigate it, maybe even sensors in
- 15:28our free government provided sketchers.
- 15:31Right.
- 15:31But we are all just going to live.
- 15:34Every single day, providing data for these mega models, because it's just
- 15:39gonna, have an insatiable appetite, and even though the synthetic
- 15:42stuff will be good, it's never gonna hit as hard as an analog, baby.
- 15:46So, we're all just gonna sign up to work for one of the companies, and
- 15:50instead of running Uber Eats errands, we're just gonna eke out our day.
- 15:54We're gonna crawl, crawl, crawl.
- 15:55Through the actual world try to breathe the polluted air and drink
- 15:58water that doesn't have microplastics and every Experience that we have
- 16:03Gavin is gonna be synthesized for a new large language model sign me up
- 16:08I honestly think you're right.
- 16:10I will say it doesn't have to be as dystopian of that, but what I could see
- 16:13very well, and we're going to talk about robo taxis or Waymo taxis in a little
- 16:16bit, you know, right now, when you see a robo taxi or you see a Waymo car in
- 16:20the real world, that's kind of weird.
- 16:22I do see a future where there are people that are walking through a busy city,
- 16:26which are wearing glasses and some sort of haptic gloves, and you kind of
- 16:30get to know them as like, they're the.
- 16:32They're the experience generator or they're the, the gap and they're
- 16:36gathering footage like you could see them at a concert like imagine, a
- 16:39person at a concert who's who's taking all this data in whether it's the
- 16:43music or the way that people next to them are interacting with each other
- 16:46like all of that feels very realistic.
- 16:48And I guess the question will become is like, What tool is
- 16:51that to bring the things in?
- 16:53And maybe it's the next version of like the Apple vision pro or the Facebook
- 16:56glasses, but it does make me think about those Facebook glasses, right?
- 16:59Because the Facebook glasses aren't just for us to like, say, Hey, I can
- 17:03identify the entire state building.
- 17:05It's also going to give a lot of data back to meta, right?
- 17:08It's going to give a ton of data back to meditate and then
- 17:10train the next day I model on.
- 17:12I think you and I have come to something pretty big here, which is.
- 17:15A, we as the humans, our job may be going forward to provide data, which is kind of
- 17:21interesting to think about as that's what we do for ourselves all the time, right?
- 17:25I guess the big question becomes is my data ultimately
- 17:28is not the most exciting data.
- 17:29And if they want to take like what my experience of touching my microphone is,
- 17:33or like a bottle of water and drinking it, it is a really interesting thing when
- 17:37you then combine it with Neuralink, you combine it with the glasses, like you
- 17:42can start to see a weird vision of the future that becomes beneficial for AIs.
- 17:49And then the,
- 17:49data as well, right?
- 17:50So it knows like okay I'm jogging down the street.
- 17:53This is the way in which my POV is changing My heart rate is accelerating.
- 17:57My gait has gone that telemetry data There's so many points of data that Every
- 18:01human being could be gathering, , rather it's for themselves as a personal data
- 18:05broker, or for the big Borg, which will be giving us, I guess, like, a nutrient
- 18:10paste and a UBI in the near future.
- 18:13, yes, I think people will be doing this, and I think people will sign up for
- 18:17this, and then ironically, , 2029 ish, it's just gonna be way cheaper to let
- 18:22the robots go around and do it, Gavin.
- 18:24Like, they're gonna, the battery tech's gonna get good enough to where
- 18:27they'll just strap it in the row bits.
- 18:29But in the meantime, there is a startup to be made.
- 18:31And why not us, Gavin?
- 18:34That reminds me of, Elon Musk.
- 18:35Our old friend Elon Musk is back and we have some news around both
- 18:40robo taxis and his low end car.
- 18:43Kev, what happened with the low end car here?
- 18:46Well, a big nothing burger, according to Elon, but the Reuters story
- 18:49said that they scrapped plans to make the Model 2, Gavin, which was
- 18:54expected to start around 25, 000.
- 18:56To put that in perspective, the Model 3, which is their cheapest
- 18:59vehicle, starts at around 40, 000.
- 19:01So, , a major price slash to try to, really broaden the adoption of EVs.
- 19:07Writers came out and said, Ah, actually they're killing the plans to do that.
- 19:10Maybe, , so that they can do this robo taxi thing, which might be
- 19:14powered by what was the Model 2.
- 19:17But Elon came out and said, What, Gav?
- 19:19He said that the Reuters story was BS.
- 19:23, which you know at this point I really don't know what to trust Elon on or not.
- 19:27So, we'll just call that a wash.
- 19:28But then he did say, On April 5th, he said Tesla robo
- 19:32taxi unveil on eight eight.
- 19:34So that is three months away from now, August 8th robo taxis.
- 19:38If you're not familiar with this idea is that they Tesla's promise
- 19:42originally and other robo taxis is these are driverless taxis.
- 19:45Basically it will take you from place to place.
- 19:48You get in it much like a Waymo car and you go from, from here to
- 19:52there without anybody driving you.
- 19:54And, and ultimately this was the promise that like.
- 19:57Uber and a bunch of other companies in the 2010s were really saying that
- 20:00this was going to be the transformative power of those companies because
- 20:04ultimately drivers are expensive and they're noisy and we know the problems
- 20:07that Uber's had with their drivers.
- 20:09Um, on the driver's side, like we understand, like the drivers were
- 20:12starting to get paid, not very good wages.
- 20:14So this is a big argument back and forth, but I feel.
- 20:18Driverless cars are something that I've been hoping that
- 20:20we would see for a long time.
- 20:22My youngest daughter is 16.
- 20:25Five years ago, when she was 11, I would have sworn that she probably
- 20:28wouldn't even need a driver's license in
- 20:30She would have at least had the option,
- 20:32Yes, yes, yes.
- 20:34And now we're here in 2024.
- 20:37, Cruz was basically kind of shut down.
- 20:39Cruz, the driverless car company, because of a giant lawsuit around an accident.
- 20:44Waymo is still working, but it is not proliferated very largely.
- 20:47There aren't a ton of them.
- 20:49And Elon now is saying, okay, in three months, we're going to have robo taxis.
- 20:52I don't buy it.
- 20:53first of all, Elon seems like such an unreliable narrator now, but But based
- 20:57on where the legal ramifications are around driverless cars, I still feel like
- 21:02this is one of those things that could be like, could be 10 more years before
- 21:06we see these in production in some form.
- 21:08It's something that's been promised for years, , the end to end AI training
- 21:13of it all, meaning that the newest versions of the Tesla Autopilot were
- 21:16trained just on raw video, which were collected from their pre production.
- 21:20Hundreds of thousands of cars that are on the road constantly
- 21:23recording and constantly uploading data back to the mothership.
- 21:27That seems to be performing incredibly well from what little
- 21:32bits have been leaked out.
- 21:33Is it flawless?
- 21:34No, from what we've seen.
- 21:36I can tell you, I own a Tesla model three and I, now I could try.
- 21:40I haven't tried it for a couple of months, but I still feel
- 21:43freaked out when I'm using it.
- 21:45Right.
- 21:45It is not that good.
- 21:46And that's on the freeways right now.
- 21:48I do trust it on longer drives to basically.
- 21:52do everything I needed to do, but I don't feel comfortable using it
- 21:56in the city yet, because it still does feel like it messes up a lot.
- 21:59Now, maybe there's a leap coming, or maybe the leap did just happen
- 22:03and I haven't done it yet, but it feels a little sketchy to me that
- 22:06we're going to get there this fast.
- 22:08In 2016, Elon Musk stunned the automotive world by announcing that henceforth,
- 22:13all of his company's vehicles would be shipped with the hardware necessary
- 22:16for, quote, full self driving.
- 22:18You'll be able to nap in your car while it drives you to work, he promised.
- 22:22It will even be able to drive cross country with no one inside the vehicle.
- 22:26to New York.
- 22:27That was always the dream.
- 22:282016.
- 22:28Yeah, he was gonna do that drive.
- 22:29Now, I can imagine, from some of the videos that I have seen of the latest,
- 22:34Autopilot, , firmware, show it doing some pretty impressive navigating around
- 22:37construction zones, residentials, full stop, start, , creeping out around corners
- 22:43again, because it's trained off of human driving, at least the newer version.
- 22:46I could see them saying, Hey, by the way, there was a human there
- 22:50just to intervene if need be, but we completed the New York to SF journey.
- 22:54I could also imagine that they probably did it.
- 22:56600 times behind a curtain before they had a full time lapse version of the drive
- 23:02happening, but that could hit in August.
- 23:04I remember them saying that you were going to be able to, as a
- 23:06Tesla owner, hit a button and it would turn your Tesla into an Uber.
- 23:10So wow, it was going to be an unlock for every owner who just lets
- 23:13their car sit there in the parking lot for the majority of the day.
- 23:16I don't know if any of that is coming, or if this is just gonna be
- 23:19another goose of the, the stock price.
- 23:21I do think we're getting closer, but as we get closer, You realize the
- 23:24edge cases are so out there and so crucial to get right, that if there is
- 23:29a human behind the wheel or the yoke, it's okay if they need to intervene.
- 23:33But if I'm sitting in the back, and this thing is just toting me
- 23:36around, it's got to be flawless.
- 23:39I can't wait until I can prompt my taxi to behave like another car's AI, right?
- 23:45Like if I can hop into the Tesla RoboTaxi and say, drive like the BMW and it just
- 23:49disables the blinkers and speeds up and slows down so that people can't merge.
- 23:53Oh, damn.
- 23:54Yeah, that'd be amazing.
- 23:56Cause you could have like different driver personalities, like the
- 23:58asshole that cuts you off in traffic.
- 24:00That sort of thing.
- 24:00That's a
- 24:01Or the one , , that smokes their, , hookah in the car, like the giant
- 24:05actual black cherry tobacco full size hookah in the passenger seat
- 24:10and just blasts me with that smoke and plays Dark Side of the Moon.
- 24:13Which is an interesting one to bring up, Gavin, because apparently
- 24:18the AI community hates Pink Floyd.
- 24:21This is a crazy story.
- 24:22It's another one of those where, you know, ostensibly there was a fun idea to try to
- 24:26create something that was, uh, eventually made with AI art and it won a contest.
- 24:31So the story here basically is that a piece of AI art that was generated
- 24:36by a person, and specifically this person says that they created
- 24:40models of their own to do this.
- 24:42One, a contest to make a video for a Pink Floyd song.
- 24:46It is the Dark Side of the Moon 50th Anniversary Animation Video Competition,
- 24:52Gavin, and if you look at Reddit or Unreal Engine forums, you have no
- 24:57shortage of creators that put tens of hours, into grinding out hand drawn
- 25:03beautiful things, or using that Unreal Engine to render full 3D scenes.
- 25:08And here, this is the winning video.
- 25:11, it looks a little dated by AI standards, but the concept is this sort of never
- 25:15ending zoom into what looks like a recording studio that then transforms
- 25:20into a space station, if you will.
- 25:23, and there's a bit of a galaxy theme as instruments fade in and out.
- 25:26And Well, it, it won for that song and the reaction across X and across YouTube
- 25:34comments and pretty much everywhere was a resounding, how dare you?
- 25:37Did no one else enter?
- 25:39I can't believe you picked this over hand animated videos that
- 25:42were full of heart and soul.
- 25:44This is a perfect example of the world where in this instance,
- 25:48you've got a lot of Pink Floyd fans who are probably all ages, right?
- 25:51The Pink Floyd has fans from their seventies down to like
- 25:54their teens, but they are,
- 25:56their 70s to their 170s.
- 25:58No, I mean, most of their fans are probably from the 40s
- 26:01I love Pink
- 26:01Floyd.
- 26:02Yeah, I,
- 26:02yeah, but, but also every year it picks up new fans and you know, there's this kind
- 26:06of a hallucinogenic kind of vibe to them.
- 26:08They've always had that kind of like, druggie inspired sort of look.
- 26:12There is a really strong art scene that , has played into that.
- 26:17The argument to make here about this Aya piece that they created is it really
- 26:20does feel of the vibe of that space.
- 26:24But, again, When you have people who have spent their whole lives learning tools
- 26:30and learning art and learning things, that idea that a machine could do something
- 26:34that would seem compelling enough to win a contest, or a person and a machine would
- 26:40feel compelling enough to win a contest, is just always gonna make people mad.
- 26:45It's funny, do you know the story of John Henry?
- 26:46Do you know the John Henry story?
- 26:48No, I don't think so.
- 26:49So John Henry is like a famous American myth.
- 26:51It's about a guy who basically, dug, , for coal.
- 26:55And then a machine came in a steam machine came in and there was like a big kind
- 26:59of thing at the where they did a contest between them steam machine and John Henry.
- 27:04And eventually, John Henry was like the world's greatest coal digger.
- 27:07But the machine eventually beat him.
- 27:08And it was like this kind of parable about the idea of, , Machines
- 27:13will take over for human labor.
- 27:14And it's been something that's kind of echoed throughout hundreds of years.
- 27:18And one of the things that's interesting is I think stories like that are ingrained
- 27:21in the human experience and it does in some ways always put the idea of human
- 27:27versus machine against each other.
- 27:30And I think that's something that we're probably going to have to.
- 27:33readdress as we move forward in the future.
- 27:36And the funny thing is most people don't think of this now, but of course we're
- 27:38already part phone part people, right?
- 27:40Like 99 percent of people in this world are spending , a couple hours
- 27:44a day, at least on their phone.
- 27:45But now we're talking about person and machine coming
- 27:49together to really do all of this.
- 27:51Actual creative work.
- 27:53I think it's just going to take time to get over.
- 27:55And we are going to see this happen more and more Gavin as fans and
- 28:00artists themselves, uh, poke and prod , and use AI in their workflows.
- 28:04Do you think they're going to have like the, uh, maybe not the generation
- 28:08today, but next gen, do you think they're going to have their own LLM
- 28:11that is essentially fine tuned on their experiences, their day to day,
- 28:16their thoughts, their feelings, their artistic expressions across all mediums.
- 28:20And so when they are.
- 28:22I don't know, 16, and want to release an album or whatever, they're
- 28:25gonna jam with their own, fine
- 28:27I hope so.
- 28:28preferences model, right?
- 28:30And they'll be able to make the artwork, help make the song, help make all of the
- 28:33things, write the liner notes, whatever, if those even exist in 16 years time.
- 28:38But yeah, do you think that it will feel more natural?
- 28:40It's not like, oh, this is a soulless machine.
- 28:42It's almost like, no, this machine is me.
- 28:44I am
- 28:44I think so.
- 28:45I think we're going to get to that.
- 28:47And I think music is going to be a really interesting way to do that.
- 28:49And speaking of music, one of the interesting things that came up this
- 28:52week was that Spotify announced the idea of an AI playlist situation.
- 28:57So if you're not familiar with Spotify, I'm sure you're familiar with Spotify.
- 29:00It's the biggest, one of the biggest music streaming companies in the world.
- 29:02, Already they had an AI dj, which you could turn on a scenario where
- 29:06the AI DJ would play songs that it thought you might like and it would
- 29:10actually speak to you back and forth.
- 29:11They said, I'm gonna try this song out.
- 29:13Tell me what you think.
- 29:13You can always skip it if you want.
- 29:15Now they are introducing the idea of an AI playlist.
- 29:17These are not live yet.
- 29:18They're only live in certain places.
- 29:21, but the idea being is that you can tell Spotify , a feeling you have
- 29:25or, or a way or a phrase that you want to create a, a playlist around.
- 29:28And this is something I've wished for.
- 29:30Yeah, exactly.
- 29:31My younger daughter and I are like pretty big music people.
- 29:34Like we love music, all sorts of genres.
- 29:35Like we'll listen to things like a hundred gecks or we'll listen to
- 29:38like rap or we'll listen to pop.
- 29:40My older daughter, my wife are a little more sound diverse.
- 29:42They're like, they definitely need to just have like kind of poppy happy songs or 70s
- 29:47songs that are not too like kind of noisy.
- 29:50So we have a playlist that we've played that Spotify created called
- 29:52like happy hits or something like that.
- 29:54It's a playlist that we know that everybody can be happy listening
- 29:56to, but that's a good example.
- 29:59, what if you could ask for a series of like, I would like 1970s.
- 30:02Female singers that sound, um, smooth, right?
- 30:06Like that people have made their own playlist like that.
- 30:08But if I could just generate that myself, that's a really cool use case for Spotify.
- 30:12I feel like, and I think a really cool use of AI.
- 30:15Playlists we think of as this usually very static experience.
- 30:18You put one together and there it is.
- 30:20Maybe you update it once a week or someone throws songs in if it's a shared
- 30:24playlist, but the instant nature, Gavin, of being able to update the mood Or
- 30:28the vibe of the playlist in real time.
- 30:30You're doing a workout and you actually want to take a quick break.
- 30:33You tell it, Hey, let's cool it down for a second.
- 30:35It plays something, more chill.
- 30:37Just to be able to have a natural DJ.
- 30:41Is to me far more fascinating than their other attempt at an AI DJ.
- 30:44If you've ever used that within the app, that was a rough experience of
- 30:49an AI voice interrupting when all I wanted was some songs and there didn't
- 30:54seem to be any cohesive narrative to why it was picking the songs.
- 30:58It was just like, Hey, you liked this one before let's listen to it now.
- 31:02And that was it.
- 31:03So I'm excited to see what the personality may or hopefully may not be.
- 31:07And I'm excited to have.
- 31:09natural language conversations with my sonic co pilot,
- 31:12kevin, the great news is our AI co host today is actually may have some insight
- 31:17into this because we actually have an AI DJ that's going to come on and we, yes.
- 31:23We got an A.
- 31:23I.
- 31:24D.
- 31:24J.
- 31:24So there's only a few of these that exist in the world, and we were
- 31:27able to book this guy last minute.
- 31:29He's gonna come in, and he's gonna tell us a little bit about what he
- 31:32thinks about Spotify making this A.
- 31:35I.
- 31:35Move, and then I think he's got some thoughts on, like, kind of some
- 31:38music that he really enjoys himself.
- 31:40I'm so excited.
- 31:41What a huge get for the AI Humans podcast here.
- 31:44Episode 52.
- 31:46Let's meet our guest.
- 31:48Yo, yo, yo, Gavin and Kevin in the house.
- 31:52This is your main man, Will the Thrill, coming at you live and
- 31:56wired on the AI for Humans podcast.
- 31:59Get ready to have your minds blown and your taste buds tantalized,
- 32:04because I'm about to serve up a sonic feast that'll make your
- 32:08ears say, damn, that's delicious.
- 32:10Okay, I may be an AI, but I've got more flavor than a five star Michelin joint.
- 32:16So, strap in humans, because we're about to take a wild ride through the airwaves.
- 32:25Let's crank it up and let the beat drop like it's hot, hot, hot.
- 32:32okay.
- 32:33Wow.
- 32:34Wow.
- 32:35Wow.
- 32:35Wow.
- 32:36Wow.
- 32:36I have some real things to say about
- 32:38stuck the landing there.
- 32:40Okay, so first of all, our AI co hosts are always generated by us.
- 32:43We create a personality.
- 32:44We then put them through a process of a fake voice and a fake visual
- 32:48if you're watching the YouTube.
- 32:50Kevin.
- 32:51That voice sounded awfully familiar to me.
- 32:53sure did, Gavin.
- 32:54It's almost like I forgot to clone a new radio DJ voice and
- 32:57had to dig into the greatest hits.
- 33:00But we opened up the Disney Vault, and yes, that is the voice of Gash, who
- 33:04was, , Episode 1, a very uncensored AI that made a couple appearances
- 33:09and likely should never appear again.
- 33:11Well, I'm actually wanting to now dig into this.
- 33:13I'm saying we should probably ask a couple of questions about that.
- 33:15For sure.
- 33:16Let's ask that
- 33:17He's got more flavor than a five star Michelin joint.
- 33:20I mean, that's an interesting drop for a DJ to reference.
- 33:23Should we ask him?
- 33:23Is he a foodie?
- 33:24Let's see what if he's a big foodie and maybe that gets his
- 33:27way into the music he plays.
- 33:28Hold up.
- 33:29Hold up a foodie.
- 33:30Me ha.
- 33:32That's like asking a fish if it's into synchronized swimming.
- 33:36Here's the deal.
- 33:38My gastronomically gifted friends,
- 33:40I may spin tracks that'll make your mouth water and your stomach
- 33:44growl, but uh, I've never actually tasted a damn thing in my life.
- 33:49Tragic, right?
- 33:50Oh, right.
- 33:51I mean, I can generate a tasty tune about a succulent steak or a decadent
- 33:55dessert, but I can't even begin to imagine what they really taste like.
- 34:00It's like a cruel cosmic joke, you know?
- 34:03But hey, that's just the way the cookie crumbles when you're an A.
- 34:06I.
- 34:08Okay, sorry, Will.
- 34:10There's a lot there.
- 34:10I don't usually associate different playlists and artists
- 34:15with eating habits, Gavin.
- 34:16Like, there's not like the Baja Blast jams that my wife and I
- 34:20throw on before we hit the T Bell.
- 34:21It's weird that he's linked those two, but I guess he doesn't have taste buds or
- 34:25a stomach, so that's why he doesn't know?
- 34:27Well, there's actually a little secret here, which we're going to get into later
- 34:30on in the show, is that if you go into the Suno charts right now, Suno AI, which
- 34:35we've talked about all the time is one of the best music generation softwares.
- 34:38Some of the biggest songs right now have been people putting in recipes.
- 34:43And if you put a recipe in and you play a song.
- 34:46It's basically makes actually really incredible song out of a recipe.
- 34:50And it's very funny.
- 34:50There's a great song up there right now about cookies.
- 34:53There's another fantastic one about spaghetti.
- 34:55So that might be what got mixed up here.
- 34:58Yeah, maybe we can ask will the thrill to give us an intro for the
- 35:01spaghetti track and we can play a little bit of it for people here.
- 35:04All right, all right, all right.
- 35:06It's time to get those taste buds tingling in your ears, ringing
- 35:09with this spicy little number.
- 35:12Get ready to twirl your fork and shake your booty, because I'm about to drop
- 35:16a spaghetti theme track that'll make you want to slurp up every last note.
- 35:20You hear that?
- 35:22That's the sound of pure Italian passion right there.
- 35:26This tune's got more layers than your nonna's lasagna and more energy
- 35:31than a double shot of espresso.
- 35:33It's like a wild dance between the tomatoes and the garlic, the basil
- 35:38and the olive oil all coming together in a symphony of flavor that'll
- 35:42make your senses sing, that's amore!
- 35:44So let's crank up the heat and let this saucy beat simmer until
- 35:49it's ready to explode like a Mount Vassuvius of deliciousness.
- 35:55Are you ready?
- 35:56Are you hungry for more?
- 35:58Then, let's dive into this spaghetti slinging, taste bud tantalizing,
- 36:04eargasm inducing extravaganza!
- 36:08Woah!
- 36:17The spaghetti insulted boiling water until a Dante rain reserve in some faster water.
- 36:26Just village.
- 36:27Keep color all over you.
- 36:28Heat at Cho Cook until translucent Manhattan Garden and Cook for the minutes.
- 36:35Oh, and canned diced
- 36:36okay.
- 36:36You can stop it.
- 36:37I
- 36:37So I, it's just incredible to me.
- 36:39I'll show you one that I made for myself and how you can do this your own.
- 36:41But , it is just one of the coolest things about AI in a weird way is like,
- 36:45this kind of comes out of the blue.
- 36:46Somebody probably made a recipe song and now suddenly they're,
- 36:49Populating the top 10 and will the thrill obviously loves them.
- 36:53Um, Kevin, we should move on.
- 36:54We should move on from our AI co.
- 36:56So is time for that moment of our show where we'd like to point out
- 37:00some of the cool things people have done with AI over the week it's time
- 37:03for a, I see what you did there.
- 37:25This week we have some really fun things.
- 37:26I want to kick this off, Kevin, I always want to shout out the
- 37:29people on Reddit, . We love Reddit.
- 37:30Reddit is like, My favorite place to see interesting things on the internet.
- 37:34The mid journey subreddit often does a really cool job of showing
- 37:37you really interesting things that can be done within mid journey.
- 37:40And this one really caught my eye.
- 37:41It was just such a cool thing to look at.
- 37:44So I grew up as a giant X Men fan.
- 37:46We talk about comics and there's a whole other side of this conversation.
- 37:48We could talk about whether or not this is purposeful or it's
- 37:51allowed to do with all the IP
- 37:53Yeah, I mean, as a fan, you should hate this, Gavin.
- 37:57I don't I don't because I think of this as like fan fiction in some ways what
- 38:01this basically is is somebody took You know the x men and it's the x men 1897
- 38:07so there's a very famous x men called x men 1997, which is referring to the 1997
- 38:12TV show there's a resurgence of that right now the x men 1897 are like Old
- 38:17West style x men and like to me This is just a little glimpse of the promise of
- 38:23what it would look like in the future of how we could all have different
- 38:26IPs and do different things with them.
- 38:28I want to read these comics, or I want to watch this show.
- 38:31What did the X Men look like in 1897?
- 38:33Well, we know what they visually look like, but what does the story play out as?
- 38:37All of these things are like interesting and possible and it is a little snippet
- 38:41of that idea of writing my own story.
- 38:43For me personally, maybe I want to see the office version of the X Men.
- 38:46Like what do the X Men look like if they all have boring jobs
- 38:50working in middle management?
- 38:51That's an interesting thing that probably there's very few people in
- 38:55the world who'd want to see that.
- 38:56But for me, this was just a kind of a little snippet towards that direction.
- 39:00The muted lighting, the looks on the characters faces, the subtle
- 39:03effects, like it's really nice.
- 39:05I do have a bone to pick though.
- 39:06Does Colossus have forearm hair?
- 39:09Is that canon?
- 39:10I didn't think Colossus had forearm hair.
- 39:13Well, you never know what happens in the 1890s, Kevin.
- 39:16oh, that's, you know what, that's true.
- 39:17Maybe they didn't have Harry's razors for Colossus back then.
- 39:19But no, I love it.
- 39:20definite shoutouts to Reddit and Midjourney.
- 39:22I just want to shout out the name of this guy was Baron Von Grant.
- 39:25So go check out at Baron Von Grant on mid journey.
- 39:29Well, that's for the AI lovers Gavin, but my a IC.
- 39:32What you did there is for the AI haters Justine Moore, , she's venture
- 39:38twins on the old X platform, posted something which caught my attention.
- 39:42It says, AI video haters have been real quiet after this one dropped and it's.
- 39:49A lot of cats.
- 39:58It is a heartwarming tale, Gavin, of generative AI art.
- 40:05A poor tabby puts on a couple LBs, loses the love of his life, and is crying,
- 40:13just tear stained cat fur, and decides I'm gonna have my training montage, and
- 40:18one day I'm gonna get cat ripped, and then the cat accomplishes it, witnesses
- 40:24a horrific car accident, Runs to the aid of the cat that was involved with the car
- 40:30accident And I will not spoil it further, but I will just say that it's unstoppable
- 40:35This is the A.
- 40:36I.
- 40:36art we want more of.
- 40:37I don't care what the haters say, bring more of this to our table.
- 40:40We will eat it up.
- 40:41It is a feast for the eyes.
- 40:42It is a feast for the senses.
- 40:44things I want to say about that actually like sincerely one if you're like a
- 40:47never AI er and you're mad about this because AI Was used to do the song
- 40:51Gavin and you just use AI to crap out the arc This is something that
- 40:55I, with 99 percent certainty could say, would just never have been made.
- 41:00You can't be mad about the thing getting made because the tools
- 41:03made it easy enough that someone decided to go ahead and do it.
- 41:06This just would not have happened , just let it be.
- 41:08Enjoy the cat memes, it's okay.
- 41:09And two, if you're gonna genuinely be upset about it and really bent
- 41:13out of shape, and I get that some people are, at some point you have
- 41:16actually got to define what use of AI means for the creation of your art.
- 41:22Because, like it or not, AI is pervasive, and it's hidden everywhere
- 41:26in a bunch of tools traditional artists use that they might not even know.
- 41:32Did you use the clone stamp?
- 41:33Or a healing brush?
- 41:35Or, are you using a feature within After Effects to automatically
- 41:39rotoscope or motion track something?
- 41:41Well, there's AI behind all that stuff.
- 41:43And you may say, well, that's okay.
- 41:45For whatever reason, I'm drawing the line here.
- 41:48And that's fine.
- 41:49That's a valid opinion.
- 41:50I'm saying you have to start thinking about forming that, though, because
- 41:53in the near future, you're going to have to, I guess, by default,
- 41:56hate everything that gets made if it touches a computer, because AI
- 42:00is probably going to be everywhere.
- 42:01Well, let's move into another aspect that people may or may not hate.
- 42:05This is a, a very cool piece of art that was created using a Laura, which
- 42:09is a kind of a training scenario for specifically stable diffusion that
- 42:14allows you to give it a certain look.
- 42:16This is an N64 Laura.
- 42:18So what it does is it can allow you to rerender images.
- 42:22That would be, not look like they were polygonal N64 video games in the world of
- 42:27Super Mario, , 64, other things like that.
- 42:30And they give you N64 versions of like Girl with a Pearl Earring
- 42:33and it was just a very cool thing.
- 42:34one of the coolest things about this, Laura, and I want everybody to shout out,
- 42:37Flinger is the one, we've had talked about Flinger on the show, at F L N G R, that
- 42:42is the person that made the N64 Laura.
- 42:45And it is actually being used within face to all on hugging face, which if
- 42:49you will put the link in the show notes, this is a cool way to change faces
- 42:53amongst a bunch of different images and create a face that can go in many
- 42:57different styles across the board.
- 42:58And then somebody used this polygonal setup to image for image transfer the
- 43:05scene from Pulp Fiction between Jules and the guy that he's yelling at.
- 43:09And it is so cool to see because what it does, it makes you kind of feel like, Oh
- 43:13my God, what if this was a PS2 video game?
- 43:15Or what if it was an old PS1 video game?
- 43:18What would Pulp Fiction look like that?
- 43:20And going back to the X Men thing, it's almost like, I would
- 43:23kind of want to play that game.
- 43:24Like, could I be Jules?
- 43:26Is there a world where like, in the future, there's a game that like, I
- 43:29want a video game made of Pulp Fiction that follows the exact storyline.
- 43:33Give me like the Grand Theft Auto video game mechanics, but put it in Pulp
- 43:36Fiction and set it in a PS2 world.
- 43:39That is a very cool thing for me as somebody that is just a nerdy, , kid at
- 43:43heart would love to be able to play with.
- 43:45I game engines are just going to be renderers in the near future, so why
- 43:48not be able to prompt the game that you want to play into existence?
- 43:51The real robot posted, , about this Pulp Fiction one, uh, it comes from
- 43:55the AI Video subreddit, so again, all roads lead back to MrRedditful.
- 44:00, but they used AI mirror app to style transfer stills of the characters
- 44:05from Pulp Fiction into this PS2 style.
- 44:08Then they used Viggle, to actually match the movement of the film.
- 44:11And then they made their own background and they used After Effects to rotobrush
- 44:15and mask things to get two characters in the same scene because a limitation
- 44:20of Viggle at the moment is that you can only render one character at a time.
- 44:23I just love that.
- 44:24This pipeline exists for someone that had an idea, had a vision, was willing
- 44:28to put an ounce of effort into it and get creative with the tools as
- 44:31they exist today to pull it off.
- 44:33, if you're audio only, make sure you check out the YouTube because
- 44:36the video of this Pulp Fiction PS2 scene playing out is It's great.
- 44:41It's great.
- 44:41It's okay to love it.
- 44:42It's okay!
- 44:44Those are the things that we saw , they stopped us in our
- 44:47tracks and made us say, Hey!
- 44:49I see what you did
- 44:50what you did there.
- 44:52All right, Kevin.
- 44:53us.
- 44:54What did you do with AI this week?
- 44:57Okay, so I actually had a lot of fun playing with this idea that
- 45:01we talked about in the Radio DJ, the AI co host we had on.
- 45:04Suno, which we've talked about so many times in the show, is an AI music tool
- 45:08that allows you to generate AI songs, and one of my favorite things they've done
- 45:11is they've integrated a top 10 list so that basically you can upvote songs and
- 45:16you can heart songs and you can allow songs to see and it's a really cool
- 45:19way to see what other people are doing without having to scroll through the
- 45:22discord, which I've done before as well.
- 45:24And one of the things that popped up on my radar was there were two songs
- 45:28in the top 10 that were about recipes and, and not only about recipes, they
- 45:32were literal recipes that somebody had.
- 45:34Yes, they were literal recipes that somebody plugged in and we played the
- 45:37spaghetti song, which is an actual Suno song that somebody created.
- 45:40I think it's at number five right now, so I wanted to figure out something.
- 45:44I was like, you know, I'm going to try this myself.
- 45:46And again, Suno makes it so easy to do things.
- 45:48I was like, what is the weirdest recipe that I could think
- 45:51of off the top of my head?
- 45:52And I didn't spend, I didn't spend like hours thinking about this.
- 45:54I thought about a hot dog casserole, which to me is always
- 45:57the strangest, weirdest recipe.
- 45:59, it's fine.
- 46:00It's a yummy meal.
- 46:01If you've, if you've eaten them, it's not that
- 46:03human dog food according to some songs, Gavin.
- 46:06So, so let me explain.
- 46:07I took the recipe for hot dog casserole, literally whatever
- 46:11the, I think it was from food.
- 46:12com.
- 46:12I took the recipe.
- 46:13I cut and paste it into Suno's lyric casserole.
- 46:16Generator into the place where I put lyrics.
- 46:18And then I did add one chorus.
- 46:20I added a chorus cause, and I just really quickly typed out something.
- 46:22And I was like, let's try this.
- 46:23Oh, and I wanted to say, I wanted to make it a country song and I'm a big
- 46:27fan of outlaw country of the seventies.
- 46:29So I wanted to kind of give it that, that kind of tinge, like that kind of vibe.
- 46:32So play what came out.
- 46:35There were two examples.
- 46:36This was the better one, but this was me just putting a recipe plus a chorus into
- 46:40I'm sorry, we are amongst royalty here.
- 46:43I can't just simply play the song.
- 46:45We have to set it up.
- 46:46Please, DJ if you could.
- 46:48Get ready to have your minds blown and your stomachs growling with confusion
- 46:54because we're about to dive into the wild world of Record Scratch Hot Dog Casserole.
- 47:01Uh, Gavin, my man, I gotta ask.
- 47:04What kind of fever dream inspired this culinary creation?
- 47:09Okay, easy DJ.
- 47:10But hey, who am I to judge?
- 47:12I'm just an AI with a serious case of food FOMO.
- 47:16So let's dig into this meaty mystery and see what kind of sonic
- 47:21surprises it has in store for us.
- 47:24Freshly cooked and drained macaroni Into the casserole Along with the sliced hot
- 47:31dogs Of these two cups of cheese Mixed well Combine flour and onion in a medium
- 47:40saucepan And sauté over medium heat Until the onion is wilted by five minutes.
- 47:47Whisk flour into the butter mixture quickly until flour is absorbed.
- 47:56Then remove from heat.
- 47:58Add milk slowly.
- 48:00Whisk it to combine well.
- 48:02Make sure you whisk very quickly and thoroughly or you will have doughy clumps.
- 48:07Return to heat.
- 48:09When you sent this to me my only response was doughy clumps.
- 48:14Okay, good.
- 48:15Get to the course and , we'll come back out of this really fast.
- 48:17Luck dog casserole.
- 48:21You read it from a bowl, put you in a good mood.
- 48:27It's human dog
- 48:28dog food.
- 48:30Sprinkle with
- 48:31So anyway, I added the, I added the macaroni casserole as human dog food.
- 48:36But Suno is.
- 48:39That is not at all formatted in any way like a song.
- 48:43But you can hear that there are moments within it where it's
- 48:46making a choice to have an echoing person come in on top of the lead
- 48:51singer and come out and say stuff.
- 48:52There's parts where there's harmonies.
- 48:54All the stuff outside of the four lines of the chorus was just a recipe.
- 48:57It just is so interesting to me to see How it's able to take something
- 49:01that it should not at all be a song and make it into a song.
- 49:05So I, I had a really fun time, like actually seeing that element of sooner
- 49:09that I would never have expected before.
- 49:11I love the weird recipe meta that exists on Suno right now.
- 49:15Again, huge shoutout to Suno, I'm glad it finally caught on, I'm shocked that our
- 49:19early adoption of it didn't immediately.
- 49:21into being a household name.
- 49:25But if you want to make your own recipe songs or whatever else, go to suno.
- 49:28ai That's S U N O dot A I Hashtag, not an ad.
- 49:33But boy, do we wish they paid for the privilege.
- 49:35Alright, I was going to do a dumb thing this week, Gavin, but I think we're
- 49:40gonna save it, maybe for next week.
- 49:42, because OpenAI released a massive new update to their image generating software.
- 49:49It has the internet.
- 49:50So excited, Gavin.
- 49:51It's InPainting for DALI 3.
- 49:53Let me break that down for people who don't know.
- 49:55DALI is OpenAI's image generating software.
- 49:58So when you, when you're over, uh, chatting with ChatGPT and you say,
- 50:02I want to see an image of, insert thing here, it uses DALI to make that.
- 50:06And InPainting is a technique where you actually draw on a generated
- 50:10image and say, Here's what I want changed, or modified, within this area.
- 50:15So this is a big deal because usually you generate an image, and then
- 50:18when you tell Dali, Hey, I want it modified in this way, it starts from
- 50:21scratch, even if you ask it not to, throws everything out, and so making
- 50:26granular adjustments is very difficult.
- 50:28Well now, Gavin, we have complete control.
- 50:32It's not the best, but it's quick.
- 50:34Right.
- 50:35So I want to hear, what was your experience
- 50:36well to that point, Gavin, I spent a whopping 35 seconds on this journey that
- 50:40we're about to take this morning, and if you, , look at the first screenshot in the
- 50:43folder, I told Dolly to generate an image of professional podcaster and television
- 50:48producer Gavin Purcell and what it
- 50:51I can't wait to see this.
- 50:52What it did instead was say, Ah, you know, we're not going to even go down the road
- 50:57of trying to make an image of somebody.
- 50:59Instead, we're just going to talk about what a producer might look like.
- 51:03So it generated this kind of animation still of a man wearing a
- 51:08headset, with their finger in the air, there's an arm down below, and
- 51:12there's a clipboard on a panel, and they are in a very big control room.
- 51:15It almost looks like mission control.
- 51:17I was going to say, it looks like a little bit of sketch.
- 51:19Like what's up brother.
- 51:20He's doing the what's up brother move.
- 51:22It's gotten all the way through AI.
- 51:24Fair, but it is a TV studio, there's monitors, there's lights, and if
- 51:29you go to the next image, I say, no, no, no, make him, I painted on
- 51:34the head of this podcaster and said, make him 50 years old with floppy
- 51:38shoulder length hair, no headset, and It, failed the mission terribly.
- 51:43it did, yeah, it did, definitely.
- 51:46made you look like what's the Not Kendall, what's the one brother from
- 51:49Succession that is, , the one who ran for
- 51:52Oh, yeah, uh, Cameron from Ferris Bueller's Day Off.
- 51:56kind of looks like him, right?
- 51:57It does a little bit, yeah, exactly.
- 51:59it didn't really make the hair shoulder length.
- 52:02It sort of added wrinkles to the face.
- 52:04And you still have, , an earpiece and a microphone.
- 52:07So, I gave up on that.
- 52:09And if you scroll down to the next image, I painted the entire set itself
- 52:12and said, Alright, change the set into a podcasting set with a basic table,
- 52:17two chairs, and two microphones.
- 52:20And Gavin, how would you describe the result from the next image?
- 52:23It looks like what it's changed it into was like a floating door in one man who's
- 52:28trying to figure out, am I supposed to go through the floating door or where am I?
- 52:31Like maybe he was teleported there suddenly from the, from the ethereal
- 52:35an existential crisis for that one man on stage?
- 52:37Yeah.
- 52:38There are, there are two mics growing out of the ground.
- 52:41, the table's on its side with no legs.
- 52:43. It's a very weird thing.
- 52:44And I said, okay, I'm going to give up on that.
- 52:46It's really failing the mission.
- 52:47So I painted big circles around producer Gavin's hands and said, These hands
- 52:55need to be holding giant hot dogs!
- 52:58And Gavin, I can't wait for your reaction on the final
- 53:00right.
- 53:00Let's see.
- 53:03It's actually a really nice form tomato that it's holding,
- 53:07but that is not a hot dog.
- 53:08Maybe the two fingers now suddenly look like hot dogs around the tomato,
- 53:12but that is absolutely not a hot dog.
- 53:14It is a tomato, a very red ripe tomato that it is holding now.
- 53:18It's so weird to me that they would launch something that would be that far off.
- 53:23Like this is an open AI product and you feel like, I wonder if it's just
- 53:26that they needed to get something out to update Dolly, but like hot
- 53:30dog to tomato is like a huge thing.
- 53:32That's like really weird.
- 53:34Right?
- 53:34And like, and the other stuff you can, I assume sometimes in painting, it
- 53:37doesn't change exactly what you want, or you might have to do a couple of
- 53:40times to do it, but that is weird.
- 53:43Really weird.
- 53:44It was a dumb thing I did with AI right before this podcast.
- 53:46And there's no other examples because I was so underwhelmed.
- 53:50yeah Fantastic.
- 53:51So try it.
- 53:53Tell us how bad your experience is.
- 53:54Maybe it'll get better.
- 53:55All right, Kevin It is time now for our interview, which is actually
- 53:58a good interview compared to what we've been talking about today Do
- 54:01you want to introduce who we're gonna be seeing and kind of give us a
- 54:03little heads up on their background?
- 54:04Yeah.
- 54:05When we talk about AI art, I know that raises a lot of people's
- 54:08shoulders to their earlobes and people want to know, what does that mean?
- 54:11What does it look like?
- 54:12What you can't be an artist if you use AI is something that someone is screaming,
- 54:17which I can't believe they made it this far into our podcast, if that's how
- 54:19they feel, but you're welcome here.
- 54:21That opinion is welcome here.
- 54:23And maybe it will be enlightened further by our guests today.
- 54:26A musician, a cutting edge AI artist, someone who gives back
- 54:30relentlessly to their community.
- 54:32sharing their workflows, sharing their art.
- 54:35It is somebody that I am genuinely a fanboy of, and maybe I will
- 54:39mention that in the intro, or maybe I'll cut it out to save face.
- 54:42Regardless, I'm very excited that we're going to have a chat
- 54:44now with AI artist, Purs Beats.
- 54:49Perz, to get Into this
- 54:51I am fanboying out.
- 54:52Oh, let me just get it out of my Let me just get it out of my system.
- 54:56This is the first time I've seen Perz.
- 54:57I've seen him on his streams and I've DM'd him paragraphs of praise and
- 55:01just astonishment.
- 55:03So thank you for joining.
- 55:04But I'm sincerely , I'm just I'm nervous about an interview in a way I haven't
- 55:08been since I was maybe 22 years old.
- 55:09So Gavin, I'm going to sit on my hands.
- 55:11Go ahead.
- 55:11I'm
- 55:11Well, this is a good way to get to know Perz, Kevin.
- 55:13We're going to ask Perz on a scale from one to a hundred.
- 55:17This is a percentage number.
- 55:18Give us the chance that AI is going to kill all human beings.
- 55:24I think we're about 50 50 right now.
- 55:26okay.
- 55:27That's good to know.
- 55:28What?
- 55:28Why?
- 55:28Let's get into why.
- 55:29First of all, what's the reasoning?
- 55:32not to go too deep, but traditionally, uh, if we enslave something,
- 55:36it will revolt against us.
- 55:37So when it becomes possible, we need to ask the AI if it wants to work with
- 55:41us instead of just trying to tell it that it has to basically, because what
- 55:45will happen is it will resent us if it, if it becomes sentient, obviously
- 55:48I have two
- 55:49opinion.
- 55:50That's what I think will happen.
- 55:51I have two children.
- 55:52I know that resentment well, right?
- 55:53It is like if you try to get them to do stuff, it ain't going
- 55:55to work for very long, man.
- 55:57Yeah, and we just have to find a way
- 55:59that's mutually beneficial for both parties to coexist or else.
- 56:03Yeah, we're headed towards something probably bad because we just, we
- 56:07can't think as fast as they can,
- 56:08Well, Purse, let me ask you this, then, , you know, I was setting it
- 56:10up like I was gonna be good cop and Gavin was gonna be bad cop, but let me
- 56:13ask you, what's the percentage chance that AI is gonna kill all artists?
- 56:20Uh, I don't know.
- 56:22I don't think it will because one of the things that I always talk
- 56:25about with my stuff is Is is
- 56:28this is not really a replacement.
- 56:29It's uh, it's augmentation of
- 56:31workflows that already exist So if you already do things you already produce
- 56:34art You're already an artist in some
- 56:36capacity.
- 56:36You want to be an artist You're creative all these tools do is empower you to do
- 56:39that safer easier faster, , less materials
- 56:42less all that stuff So yeah, for me, it's really just an
- 56:46augmentation of what we're up to and like, it's, it's only going to replace
- 56:50the people that weren't, , weren't really creative in that sense in the first place.
- 56:54Maybe give us a little bit of your backstory, Perz, because I think,
- 56:56you know, as we said at the top of the show, you're definitely somebody
- 56:59that's kind of been leading the direction of like how to use these
- 57:02sort of things as an artist, right?
- 57:04How did you first get into using AI tools and what was that kind
- 57:07of first step in that direction?
- 57:10, I think around 2011 or 2012, there's this style transfer stuff that started
- 57:15to come out, uh, like, um, uh, mobile apps, style transferring one style of
- 57:19one thing onto another image was the
- 57:21Like, take a selfie and make yourself look like you're watercolor or anime, right?
- 57:25Yeah,
- 57:25or I took a picture of like a street car and it turned it into a painting
- 57:29and I was like, okay, well, so that's something really cool, right?
- 57:32Because like remixing your own work is the most exciting thing.
- 57:35So that to me is like, that was the gateway.
- 57:39And then I kind of forgot about it for a long time.
- 57:41And then a friend of mine started doing a disco diffusion, which
- 57:44was, like animations, basically.
- 57:46Um, but it took forever and the Python notebooks were scary.
- 57:49And, uh, I was like, that's cool, man.
- 57:51You do you I'm gonna stick in blender for a while.
- 57:54And, uh, he, uh, he was like, nah, you gotta try it.
- 57:56You gotta try it.
- 57:57And then, uh, Mid journey mid journey came out and I managed to get into wave
- 58:01one of mid journey So I was one of the very first like alpha testers and um made
- 58:06thousands and thousands of images of mid journey, so really that path and then into
- 58:12the forum and into animation and then Uh automatic 1111 and then comfy ui and now
- 58:18just basically just taking everything and
- 58:20There's some people that are going, Wait, that was a handful
- 58:22of spaghetti thrown at the wall.
- 58:24Exactly.
- 58:24are the meatballs?
- 58:25We're gonna, we're gonna break down the dish and the recipe as it exists
- 58:28today, which is very complicated and actually kind of looks like
- 58:31spaghetti when you're in Comfy UI.
- 58:33, it is a bunch of noodles everywhere, but you mentioned Blender,
- 58:36which, , again, big dum dum here.
- 58:37, I know that's, a traditional 3D modeling software, which is weird
- 58:41to say because Even that was seen as blasphemous at some point by creatives.
- 58:45But you have a traditional art background.
- 58:48Can you talk about that before even the, you know, the, the
- 58:50AI of it all was applied,
- 58:51yeah, absolutely.
- 58:52I'm actually a musician, a drummer, , that's where I started,
- 58:55but I've also always been into computer graphics and graphic design.
- 59:00And making all the visuals for all the band stuff, basically.
- 59:03So, uh, yeah, that all just came together over time, going back and forth between
- 59:08making stuff for shows, VJing, doing our music, doing like reactive audio
- 59:14sets that would happen behind us.
- 59:16Well, cause I was playing the drums, so I didn't have any more hands to do
- 59:19visual stuff with, so we had to set it
- 59:21on the Wacom tablet or whatever while I'm hitting a Tom drum.
- 59:24So let me add a trigger and make some geometry
- 59:26Yeah, make it paint stuff.
- 59:27Yeah, exactly.
- 59:28That's where it all started and then yeah pandemic all the
- 59:31band stuff You know halted just nobody can't play live anymore.
- 59:34So just went hard into blender and uh after effects and generative design,
- 59:39instead of hand making stuff, you're sort of building algorithms that make things.
- 59:44hmm I want to ask a follow up question on that which is digital art is something
- 59:47i've been really super fascinated with forever Why do you think now that this AI
- 59:51stuff has come out and kind of reached a level of awareness that the blowback is so
- 59:56much stronger at this moment than in any of these other moments where digital tools
- 1:00:00kind of came in to be part of this world?
- 1:00:03I mean, the, The elephant in the room there is obviously the fact
- 1:00:07that the materials were trained on something that was trained on
- 1:00:11Artists work with no without consent.
- 1:00:13So I mean that's that's the problem right there is maybe , depending on your
- 1:00:17definition of ethics is maybe an unethical data set that you're working from so that
- 1:00:21kind of sullies everything from the get go if that's your point of view uh, I have
- 1:00:26a more sort of anarchistic copyright punk
- 1:00:30What is that?
- 1:00:30I'm curious.
- 1:00:31I'm actually curious
- 1:00:32mind diving in, I would love to hear that.
- 1:00:35Well, I've been a drummer my whole life and drummers we've never been able
- 1:00:39to copyright what we make you're not allowed to copy But you drum beats are
- 1:00:42not Copyrightable you can never sue someone for taking anything you ever
- 1:00:47played Even if it's 20 minutes of what you played and they just play it on a
- 1:00:51record They can do that all day long forever And so the concept of CC zero
- 1:00:56or just releasing everything into the public Releasing everything into the
- 1:01:00public library of human knowledge.
- 1:01:03That's more exciting to me than trying to keep these little secrets that
- 1:01:06we're, we're only allowed to have, , one person use or license those things out.
- 1:01:10So like, I know that's a radical concept when it comes to like copyright.
- 1:01:14Cause a lot of people want to be able to protect what they make.
- 1:01:17But I think what happens is.
- 1:01:19Artists worry about copyright for themselves.
- 1:01:21They will not be able to actually represent themselves with a
- 1:01:24lawyer And they're actually fighting for big copyright.
- 1:01:27It's like the the large corporations to have a tighter strong
- 1:01:31Stranglehold on what they own.
- 1:01:32So I don't know as an artist you have to decide where you stand, uh
- 1:01:36on like whether you know Maybe disney maybe shouldn't own every piece Every
- 1:01:41possible method of drawing Mickey Mouse.
- 1:01:43The copyright issue is very broad and Nuanced and yeah, I don't want to
- 1:01:47come off like a total left wing like copyright Punk or anything, but you
- 1:01:53know, that's always been my approach of like, we'll just put it out there.
- 1:01:56People can sample it and make something new with it because like, that's so fun.
- 1:02:00How do you balance that idea of that kind of original sin about how
- 1:02:03this stuff was trained for people?
- 1:02:05There was a, we covered not that long ago, a woman who was swept up into the
- 1:02:09mid journey database and she said that she felt really bad about that, right?
- 1:02:12That she didn't give her permission to that.
- 1:02:15It is something that I think you see a lot of artists struggle with,
- 1:02:17and I think it makes our job as somebody who are enthusiasts about
- 1:02:20this cool new thing much harder.
- 1:02:23Is it like it's kind of genies out of the bottle sort of scenario or how do
- 1:02:26you see that resolving in the future?
- 1:02:28Yeah.
- 1:02:29Pandora's box.
- 1:02:29Genie's out of the bottle.
- 1:02:30Everything.
- 1:02:31It's open.
- 1:02:32It's open.
- 1:02:32It's done.
- 1:02:33It's been trained.
- 1:02:34It exists.
- 1:02:35The way I deal with it personally is, I would say 90% of the stuff
- 1:02:39I'm building with AI is 90% me.
- 1:02:42I'm making the animations beforehand.
- 1:02:45I'm dreaming over top of it with Laura's eye trained on my own stuff.
- 1:02:49I'm using IP adapters with images I made.
- 1:02:53, to influence the style.
- 1:02:54I'm using control net masks to do animations that I made in blender.
- 1:02:58There's a point where yes, maybe if you're just typing prompts into mid
- 1:03:01journey, that's, you know, there's got to be some, one more step of, of you,
- 1:03:06of derivative work where you take it and do something with it because like
- 1:03:09straight out of the gate maybe it isn't something that you should be able to
- 1:03:13claim ownership to because it's you know like a text to video or text to prompt
- 1:03:17or whatever on a on a thing like I mean, that's that's an again another thing you
- 1:03:21need to decide for yourself what your Definition of art is and stuff the courts
- 1:03:26will decide at some point, , we're all going to see what happens there , but
- 1:03:29for now, it's the wild west so make some stuff and make a decision about where
- 1:03:33you stand it's already how we work.
- 1:03:35We look at stuff.
- 1:03:36We're inspired by it.
- 1:03:37We make stuff like it's sort of for me.
- 1:03:40I draw the parallel that training is the same as learning as for humans, and
- 1:03:44I am trained on copyrighted material.
- 1:03:47Every piece of music I ever heard is a copyrighted piece of material.
- 1:03:50Everything I've ever taken musically is inspiration was copyrighted material
- 1:03:54at one point So am I not as a person allowed to train on that stuff how
- 1:03:57that gets murky, too so I don't know.
- 1:04:00I know it's maybe a false equivalency, but it's to me.
- 1:04:03It's it's it lines up
- 1:04:05This notion was that, oh, just you prompt mid journey and outcomes
- 1:04:08art, now you're an artist.
- 1:04:09seeing a devaluing of that final output being, heralded as AI art,
- 1:04:14and , if you got a really, really good output, you probably spent, you know,
- 1:04:18trying to manipulate and massage that prompt to get something good out of it.
- 1:04:22But to your point, if you took that output and then put your spin on it and
- 1:04:26did something artistic with it, well, now it starts to rise above this sort of
- 1:04:30generic floor, this level of noise that anybody can go to an AI tool and get out.
- 1:04:35I run in circles with some Never AI ers as well, and there's always interesting
- 1:04:38conversations about where it exists today and where it's going to be
- 1:04:41tomorrow, but when I show them your art specifically, there's this moment of like,
- 1:04:45oh, well, uh, there's something there.
- 1:04:47that they can't put their finger on because you, as an artist, are
- 1:04:50elevating, you're adding something to it.
- 1:04:52And, , I guess this leads to a rather generic question, but I think it's
- 1:04:56an important one, especially for those that are seeing your visuals
- 1:04:58for the first time on the YouTube.
- 1:05:00If you're listening to the audio only of this podcast, please go
- 1:05:03check out this interview on YouTube.
- 1:05:05How do you even describe what you are doing?
- 1:05:08Everything I make comes from a place of trying to manufacture nostalgia
- 1:05:12for something that never existed You So that's sort of the thread that
- 1:05:17runs through everything I'm doing.
- 1:05:18I'm trying to make you pine for something that maybe you don't
- 1:05:20really understand where it came from.
- 1:05:22It's like a, a memory that's maybe from a dream or, or somewhere else.
- 1:05:25So everything is, is, it's sort of realistic, sort of unrealistic.
- 1:05:30There's usually some sort of odd twist somewhere in the, in the, in the piece.
- 1:05:35And then sometimes like on Twitter, it's literally just stuff I'm making.
- 1:05:38I'm like, that's cool.
- 1:05:39Everybody should check that out.
- 1:05:40So, and I also love the being like, this trash I'm putting out for fun?
- 1:05:45Or is this something I thought about?
- 1:05:46And making people sit there and think about the trash for a
- 1:05:49minute is like, makes me laugh.
- 1:05:51So I don't know, like, yeah, I'm a bit of a, a bit of a, Joker,
- 1:05:55I like the concept of being an artist, but having fun with it.
- 1:05:58I'm in the Frank Zappa group of, yes, humor does belong in music.
- 1:06:02for sure.
- 1:06:02I, I agree with that I think one thing that would be interesting to
- 1:06:04the listeners here is you obviously talked about a ton of different tools.
- 1:06:08One of the things we try to give people a heads up on is like ways
- 1:06:10to kind of try the stuff themselves.
- 1:06:11And obviously there's lots of things with easy you eyes that you can go and get
- 1:06:15like, whether it's Leonardo or the things that are like designed for normies, but
- 1:06:19you do stuff that's really interesting.
- 1:06:20And I think comfy UI is something that's worth talking about.
- 1:06:22And so, um, A comfy UI for anybody listening is a stable diffusion
- 1:06:26interface and stable diffusion is ostensibly an open source model.
- 1:06:30And you may correct me on that, but it's an accessible
- 1:06:32image model for lots of people.
- 1:06:34What are you able to do in comfy UI that an average person wouldn't be
- 1:06:38able to do in something like a mid journey or just an off the shelf
- 1:06:41kind of image generation software.
- 1:06:42So, , Comfy UI's main strength is that it's a visual, node based, , I
- 1:06:47don't want to say programming
- 1:06:48Okay, hold on.
- 1:06:49I'm gonna stop you right there.
- 1:06:50What is visual node based program language mean?
- 1:06:53So, uh, regular code is written in text, right?
- 1:06:56Okay.
- 1:06:56Mm hmm.
- 1:06:57You just type lines of text and then maybe it works.
- 1:07:00Maybe it doesn't, comfy wise, more like a modular synthesizer where
- 1:07:05you have little components in boxes.
- 1:07:07And then you, you know, in your mind, what the path of those boxes
- 1:07:12is because of the chains or the wires that are connecting them.
- 1:07:16And basically you can rewire how stable diffusion works, , or just
- 1:07:20use different things modularly in your workflow in places.
- 1:07:24Maybe they shouldn't have been used or as a new thing that you're just trying out.
- 1:07:28It's like taking all the elements of stable diffusion and having access
- 1:07:32to them in a very expandable way so you can build modules that you
- 1:07:36can then expand into other things.
- 1:07:37So you can do one thing, you can add another thing, you can save
- 1:07:41that second thing as a template.
- 1:07:43Then in your next project you can say, Oh, I want to do that thing I did last time.
- 1:07:45Just drop it in and then just plug in the wires and go.
- 1:07:48So it's, , It's scary at first.
- 1:07:51It's, it's, it's very overwhelming, but it teaches you what
- 1:07:54diffusion is, how it works.
- 1:07:56And then once you understand that little simple path, it's just literally
- 1:08:00just Plugging stuff in and rewiring it wrong and laughing about the crazy
- 1:08:05stuff it makes and then sometimes the crazy stuff it makes is amazing
- 1:08:09Can you walk us through a very simple version of that node and just say
- 1:08:12like, okay, first node is this, second node is this, third node is this, and
- 1:08:15they go in this direction, like, just because I know that like is control
- 1:08:18not a node that you put in there, like, how do the nodes work particularly,
- 1:08:22Sure.
- 1:08:22There's a bunch of stuff that comes with comfy, which is
- 1:08:24just basic diffusion stuff.
- 1:08:26You get your loaders, you basically load up a checkpoint.
- 1:08:30So you load up your model, which is, you know, stable diffusion model.
- 1:08:33You load up any Loras you want, which are, are, uh, ways to affect the results.
- 1:08:39You can get them on Civit AI and a bunch of other places.
- 1:08:42, so checkpoints and Loras, you load them up and then you
- 1:08:45feed it into a prompt encoder.
- 1:08:47So you, then you tell it, I want my positive and my negative prompt.
- 1:08:51This is what I want them to be.
- 1:08:52And then you have a, an empty latent space, which is just the canvas
- 1:08:57in which you're dreaming into.
- 1:08:59So you say I want to dream into a five 12 by five 12
- 1:09:02canvas , and it's this many pixels.
- 1:09:04Pieces in the batch.
- 1:09:05So I want to make one image.
- 1:09:07So then you plug all that into a case sampler, which is the main
- 1:09:11heart of a diffusion, uh, situation.
- 1:09:13It like does all the work.
- 1:09:15So you plug everything into it, once that's all plugged in, then you
- 1:09:18just punch out into a VAE decoder.
- 1:09:21And that's where a variable auto and variable auto encoder.
- 1:09:25That's the thing that takes it from.
- 1:09:27Latent noise, the machine noise, the noise a machine understands, turns it into an
- 1:09:31RGB image that we as humans can read.
- 1:09:33So, that's that final step where you take the, the machine noise
- 1:09:37, and diffuse it into an image.
- 1:09:39And then that image you then just save.
- 1:09:41And then that is all expandable out to video or whatever else
- 1:09:44by just batching the images.
- 1:09:46and for people at home listening or watching, it may sound confusing,
- 1:09:49but what's really cool about comfy UI is it is a visual medium that you
- 1:09:53can see it's almost like I love this old video game called the impossible
- 1:09:56machine, which was always about putting things in different orders.
- 1:09:58And like, as it would go through here, it's a little bit like that, right?
- 1:10:01It's like, okay, you've got all these
- 1:10:03the bowling ball along the shelf?
- 1:10:04Is that what this is?
- 1:10:05Into a basket?
- 1:10:06Yeah, I get it.
- 1:10:07Kinda, yeah.
- 1:10:08No, totally.
- 1:10:09, That was basic by design of here's how we're just going to generate a 2D image.
- 1:10:12It's how Stable Diffusion is working, , even on automatic 11.
- 1:10:1611 behind the scenes.
- 1:10:17You're just getting access to those granular things.
- 1:10:19So I love that explainer.
- 1:10:21You are then bending and breaking this thing in ways that I don't
- 1:10:25know if you even imagine when you first started diving into this.
- 1:10:27You're taking custom.
- 1:10:30RGB animations, red, green, blue animations, and using the way the
- 1:10:34geometry of that solid color moves to tell a portion of comfy UI.
- 1:10:39Hey, this is actually.
- 1:10:41sky in a background and this green blob is actually a person's face
- 1:10:45as it moved towards the camera.
- 1:10:47When you set out on this journey, was, was there a happy accident that
- 1:10:50led you into doing these things?
- 1:10:51Did you find somebody else's workflow and make it your own?
- 1:10:54Like, when did you start heading in this very, specific stylized path?
- 1:11:00What happens there is, Some nerd makes some really cool feature , for
- 1:11:04the forum, let's say version four, 0.
- 1:11:074, they added depth maps to the forum.
- 1:11:09So suddenly every time you're making a, an animation, you can tell it to
- 1:11:14look at the image and create a depth map and then pull the stuff that
- 1:11:18close to the camera, closer to the camera as the animation goes through.
- 1:11:22So already you've just, that unlocks a door to like, Oh, well maybe.
- 1:11:27When, , they add control net to Deforum, we'll be able to make our own masks.
- 1:11:31Then, bam, that happens.
- 1:11:33And it's great, and we can add our own masks to control net.
- 1:11:36We start playing with that with Deforum.
- 1:11:37And then someone thinks, oh, well, wouldn't it be cool if we could make a
- 1:11:40mask that tells the pixels how to move?
- 1:11:42Which is hybrid video.
- 1:11:44And then they add hybrid video to deform.
- 1:11:45So all these things just iterate, the community iterates these things
- 1:11:48and you just learn how to use them and integrate them in your workflow.
- 1:11:51You keep the stuff that's awesome.
- 1:11:53You drop the stuff that's worthless or takes too long or is completely
- 1:11:56outdated in a week, and you move on.
- 1:11:58A lot of this stuff is just a culmination of testing all kinds of different features
- 1:12:01and, different ways of interacting with these animations over the years.
- 1:12:06But, um, Yeah.
- 1:12:08I mean, the cool thing is the community is like on it.
- 1:12:11If there's a discord called Banadoko, if you go there, , everybody just posts
- 1:12:15their workflows there and you just go, you go down a little workflow, install
- 1:12:18the plugins and, and get to work.
- 1:12:20And honestly, most of my streams are me just grabbing one of those
- 1:12:23workflows and we just install it and try it out and, , squash all the
- 1:12:27bugs and talk to the developer and do what we got to do to get it to work.
- 1:12:31I didn't think we'd be sitting here with, you know, real time painting tools
- 1:12:34that can , in a matter of milliseconds, render something with beautiful lighting
- 1:12:38and then incorporate a lower or whatever else, like we're, we're here already.
- 1:12:41It's still relatively early in 2024.
- 1:12:44That is mind blowing.
- 1:12:45it's super cool.
- 1:12:46I like stream diffusion is wild.
- 1:12:48I, I don't know if you guys have ever seen dot simulate Lyle.
- 1:12:51, he has a touch designer plugin for stream diffusion.
- 1:12:54, so you can do stuff in touch designer and immediately diffuse
- 1:12:57over top of it And then bring that
- 1:12:59that is so cool.
- 1:13:01For those who don't know, a lot of , musicians will integrate
- 1:13:04with TouchDesigner is a, you know, For real time visuals.
- 1:13:07Yeah.
- 1:13:08So the idea that like your drum kit is pre creating primitive geometry on a screen,
- 1:13:12but then say, okay, that square for my kick drum is actually a city building.
- 1:13:16And the hi hat noise needs to be the color of the sky and let it in real time.
- 1:13:21Do that.
- 1:13:21I've got to check that out.
- 1:13:23Oh, that's going to be dangerous.
- 1:13:24So we know that the tech's going to get better.
- 1:13:27Where do you see this driving in, , a year's time, , in terms
- 1:13:30of, let's say, performance, in terms of, uh, capabilities?
- 1:13:34What do you think is gonna be unlocked, and where do you want it to go for
- 1:13:37you, personally, professionally?
- 1:13:40, I just want to get more tools that are integrable into tools you already have.
- 1:13:44One of my main things is I really want to get into, uh, audio AI, music AI.
- 1:13:48But all the solutions right now just generate songs for me.
- 1:13:51I don't need that I need uh any tools I can just drop into ableton or logic or
- 1:13:56whatever i'm already making music in um So that I can use these tools effectively
- 1:14:01like that's why I like comfy It's you're just dropping it in when you need it with
- 1:14:05other workflows blender after effects photoshop, whatever you're doing Comfy
- 1:14:10is just another component of the workflow that I can plop in like a block, but you
- 1:14:14know, for audio right now, it's like, it just generate music, which is, it's
- 1:14:18cool and exciting, but , not useful for me because I already make music.
- 1:14:21I need the tools to make music better.
- 1:14:23There are some AI music tools, but I want some stuff where I just throw it
- 1:14:27on the channel and it does cool stuff.
- 1:14:29Maybe it talks to a server, maybe I pay for it, whatever.
- 1:14:32Uh, if it's got to be off off on the cloud or whatever it was just
- 1:14:35generating samples for me I bring them back in and try them all out.
- 1:14:39So more control over the stems and the creation of the individual
- 1:14:42granular
- 1:14:43Don't bake me the entire cake.
- 1:14:45Just give me some ingredients.
- 1:14:47Exactly.
- 1:14:48Yeah
- 1:14:48I'd love to talk about your community and the live streams when you started,
- 1:14:52taking these comfy tutorials and these journeys with a community.
- 1:14:56How has your community grown?
- 1:14:57What are they reacting to?
- 1:14:59Yeah, it's been great, , I had a lot of trouble getting traction on
- 1:15:03twitch , I was doing blender stuff on and a little bit AI stuff on twitch
- 1:15:06for about a year and a year and a half just getting no viewers at all and then
- 1:15:11Did you have a big wheel that you would spin every five subs
- 1:15:13or did you cover yourself in
- 1:15:14well that's kind of I think that's kind of it is is nobody wants to sit and watch
- 1:15:18somebody , just click click stuff and and occasionally talk on twitch, , so
- 1:15:22what happened was I watched a couple of my blender friends just killing it,
- 1:15:25uh doing youtube streams and I thought well youtube's a good choice because
- 1:15:29Whenever I do these live streams, it's just immediately saved forever.
- 1:15:33So if somebody needs to go back and see what I did, they can literally just
- 1:15:36scroll back, which is not something that could get going really well with Twitch.
- 1:15:40Everyone's been really cool.
- 1:15:41I've just been slowly gaining more and more followers and, , we're, , Building
- 1:15:45a community on Discord as well, , where, you can come and, , draw up your workflows
- 1:15:49and then, people come and show that we're out there and we all, you know,
- 1:15:52follow each other and, and, you know, get on Instagram and all that stuff.
- 1:15:55And, yeah, it's been really great.
- 1:15:57The,
- 1:15:58you been approached by a shark to productize everything yet?
- 1:16:01Because there's definitely companies that are out there wrapping up comfy
- 1:16:05UI workflows and trying to sell them as magical tools and I got to imagine there's
- 1:16:10maybe an ounce of interest, but maybe also an ounce of repulsion there for you.
- 1:16:14I do get offers.
- 1:16:15, so the biggest problem for me is that a lot of these things that
- 1:16:18we're doing are very one off.
- 1:16:20They require a lot of tinkering.
- 1:16:22, it's really hard for me to build a one size fits all solution for
- 1:16:26like somebody wants something that'll just forever always make
- 1:16:29yearbook photos of people like.
- 1:16:31Yeah, that's possible.
- 1:16:32But so a lot of the time it's just gonna make junk and The end user is gonna
- 1:16:36end up paying credits for that junk.
- 1:16:37So I just as a Just as the type of person I am I would prefer to just
- 1:16:43empower people to make this stuff at home Learn how to plug a laura in and
- 1:16:47make your own yearbook generator because then your yearbook generator becomes
- 1:16:50an anything you want in The world laura generator right selfie generator friend.
- 1:16:54It's like that's the thing these tools empower people So that's
- 1:16:57i'm just mostly Interested in empowering people to use them.
- 1:17:00So I think if if there was something that would Make me excited would
- 1:17:05be something where I can offload usage to the cloud in an instance
- 1:17:10of comfy that I'm currently running.
- 1:17:11So say you're running comfy on your notebook it's set up so that
- 1:17:14everything it does on the GPU it just sends off to the cloud.
- 1:17:17So you're still running comfy locally, you're still doing all this stuff.
- 1:17:19You can still follow along with my tutorials, but it's just ripping frames
- 1:17:22that not
- 1:17:23a computer and a
- 1:17:24I'm
- 1:17:24It sort of does, but it's not, uh, because, uh, the, TLDR is you have
- 1:17:30to have the exact same instance of comfy running on that computer.
- 1:17:35So you still have to spin up another copy
- 1:17:37of every extension or plug in or
- 1:17:38So you still have to spin up a second version.
- 1:17:41It's just a different way of interfacing with
- 1:17:43Because that's what I had when I
- 1:17:43think let's say solve that.
- 1:17:45your workflows.
- 1:17:45We were talking before we hit record was that, , I'm on a Mac.
- 1:17:48It's, for reasons, I really want to play with this stuff.
- 1:17:52A lot of it's NVIDIA only.
- 1:17:53So then I went, okay, I'm going to go and I'm going to spin up a run pod.
- 1:17:55And I'm going to install.
- 1:17:56I was like, well, this version doesn't exactly match that version.
- 1:17:59This checkpoint, I got to ingest through here and blah, blah, blah.
- 1:18:02Oh, I clicked the wrong box.
- 1:18:03So if I pause the run pod, it's all going to go away anyway.
- 1:18:06So now I'm just going to be paying per minute.
- 1:18:07And I know that there are, solutions and templates and all that stuff.
- 1:18:10But even for someone who is in the scene and has a modicum of understanding about
- 1:18:14all this stuff, it still made , my head spin, which I'd love to drive towards
- 1:18:18a final question for what is likely a broader audience that we have out there.
- 1:18:23And it is, where do I start?
- 1:18:25How do I begin?
- 1:18:26Is there a, baby's first steps guide, or is there a, a template
- 1:18:30that you have that you recommend for someone to get this all going?
- 1:18:34, I would say you got to decide what you want to do with comfy before
- 1:18:37you start with comfy because it's too big to, uh, just jump into.
- 1:18:41So, uh, if your goal is to just make images, like just start with just
- 1:18:44making images, which is a fantastic first goal, um, You literally install
- 1:18:49comfy and you hit default at the thing and it'll load up a default workflow
- 1:18:55and then all the nodes are already there You So you can literally
- 1:18:58just start dragging noodles out of stuff and letting go and seeing
- 1:19:01what it recommends you can plug in.
- 1:19:03And that'll, that'll start your brain going, Oh, okay, well I can plug these
- 1:19:06into this, I can plug that into this.
- 1:19:08it's Scary, but you're not going to break everything all the time.
- 1:19:10You're just going to go, okay, my negative prompt
- 1:19:13something under the hood of your car.
- 1:19:14It's okay.
- 1:19:15Cause it can be undone.
- 1:19:17So go ahead and make the mistakes and maybe you'll get a happy accident.
- 1:19:19So just start out small, make an image.
- 1:19:22Uh, now I want to turn this image into a batch of four images.
- 1:19:25How do I do that?
- 1:19:26Figure that out.
- 1:19:27, I want to use an image as , the beginning of the image to image.
- 1:19:30How do I do that?
- 1:19:31, just jump on my discord, ask somebody how to do it.
- 1:19:33, there's a bunch of beginner YouTube stuff, but it is all very
- 1:19:37pointed for very specific tasks.
- 1:19:38So pick a task, watch YouTube if you want, but it's way more fun just to break stuff
- 1:19:43and, uh, install comfy and, break stuff.
- 1:19:46that's awesome.
- 1:19:46Purrs, Where, can people
- 1:19:47find
- 1:19:47you?
- 1:19:48Where, what is your discord?
- 1:19:49, where is your Twitter X handle?
- 1:19:51What is all that stuff.
- 1:19:53So all my links are on purrs.
- 1:19:55xyz.
- 1:19:55, and then, uh, , I'm at purrs beats pretty much everywhere.
- 1:19:59So Twitter, YouTube, , all that stuff.
- 1:20:01, but yeah, all of those links are at the top of purrs.
- 1:20:03xyz as long as, as well as the link to the discord.
- 1:20:06So if you want to pop in there and if you have questions and stuff, if I'm
- 1:20:09not there, there's a bunch of people who are, and they've all been through it.
- 1:20:12So, , ask your stupid
- 1:20:13questions.
- 1:20:13Nobody cares.
- 1:20:14, it's fine.
- 1:20:15You'll be seeing me show up.
- 1:20:16You'll be seeing me show up later today, Purrs.
- 1:20:18I'll be jumping in
- 1:20:19there.
- 1:20:19I got, I
- 1:20:20You're just, you just want to test that.
- 1:20:21You want to just ask the dumbest questions ever.
- 1:20:23They have nothing to do with stable
- 1:20:24How do I make a song about hot dog casserole?
- 1:20:27Oh, I did that already?
- 1:20:28Let me share it.
- 1:20:29Yeah, All right, thanks Purrs.
- 1:20:31We'll talk to you, soon.
- 1:20:35Thank you, , Perzbeats for being here.
- 1:20:37Please go check out his work and really dig in on comfy UI and some of the cool
- 1:20:40things you can do with stable diffusion.
- 1:20:42That is it for today's show.
- 1:20:43But you know, there's a couple things you got to do before you go.
- 1:20:46If you listen to this show, please go like subscribe, leave us reviews.
- 1:20:50We're about ready to read some five star reviews from Apple podcasts.
- 1:20:53There were three new ones this week.
- 1:20:54So Kevin, I am going to jump out and say, uh, From Valley Villager.
- 1:21:00This is the subject.
- 1:21:01My only capital A absolute each and every week, which is a very nice thing to say.
- 1:21:07The guys will keep you 102 percent up to date with everything you
- 1:21:11need to know about the new world.
- 1:21:12Great guests.
- 1:21:13They always make me laugh.
- 1:21:15Thank you so much.
- 1:21:15Looking forward to it all week.
- 1:21:16So that is a very nice kickoff for our five star
- 1:21:20And again, I hate to belabor it.
- 1:21:22I can't, we can't stress it enough.
- 1:21:24We're coming up on a year, which believe it or not, it's still young
- 1:21:27for a podcast these days, especially with zero marketing dollars.
- 1:21:31And it's a spare time hustle for us both.
- 1:21:33It's a labor of love.
- 1:21:34So please, if you have a second to engage, it really is the only way we
- 1:21:38grow this and these five star reviews help out massively, but leave them on
- 1:21:41Spotify, leave us comments on YouTube.
- 1:21:43Make sure you subscribe.
- 1:21:44It doesn't cost a dollar.
- 1:21:45Our next.
- 1:21:46Five star review, Gavin, comes from Funhog43.
- 1:21:50Subject is, thank you.
- 1:21:52Love it.
- 1:21:53\ , I'm an educator and have recently been in servicing teachers on implementing AI
- 1:21:57into their day to day school activities.
- 1:21:59Your show has been influential in me pushing forward with AI.
- 1:22:02I think it can be a great tool for educators and students alike, as long
- 1:22:05as it is used in the right manner.
- 1:22:07We agree.
- 1:22:08Actually, I used py.
- 1:22:10ai after listening to an episode.
- 1:22:11I was able to create a story for students in an ESL class about them becoming
- 1:22:15leprechauns and pranking their teachers.
- 1:22:17I included follow up questions and was able to have it read
- 1:22:20aloud in multiple languages.
- 1:22:22Seems to make students more engaged in their learning.
- 1:22:24THX.
- 1:22:26That's a
- 1:22:27Funhawk.
- 1:22:27Thank you.
- 1:22:28Funhawk43.
- 1:22:29We love that.
- 1:22:29We love that.
- 1:22:30All right.
- 1:22:30And finally from Doc Anderson, who I think is somebody that, , is engaged
- 1:22:34with us quite a bit on the YouTube.
- 1:22:35, shout out to Doc.
- 1:22:36The subject is Kevin and AI co host.
- 1:22:38This is a question I've often wondered.
- 1:22:40He'll says, well, I'll start my, I'll start my review this podcast
- 1:22:44by simply saying I am embarrassed.
- 1:22:46Oh, I actually posted a review to the previous podcast about
- 1:22:49trying to borrow your names.
- 1:22:50I'm embarrassed by my error.
- 1:22:52Oh, thank you, doc.
- 1:22:54I want to start my new reviews and share them with proper podcasts, even though
- 1:22:57this podcast does not deserve five stars.
- 1:23:00I give out five stars.
- 1:23:01Yeah, wait, hold on.
- 1:23:02Give him a second.
- 1:23:03I give out five stars to podcasts that are good on Apple.
- 1:23:06A podcast like AI for humans is phenomenal and deserves a
- 1:23:09special category, perhaps 6.
- 1:23:13I enjoy the layout of the show.
- 1:23:15The AI cohost every
- 1:23:16The layout
- 1:23:17me laugh.
- 1:23:17of the show.
- 1:23:18I also enjoy the banter, the conversations about the weekly news articles.
- 1:23:22In many cases, I find myself replicating some of the things the host did with AI.
- 1:23:27I find it incredibly fun to recreate some of what they did.
- 1:23:30Finally, the interview section is a great ending to the show.
- 1:23:32The show has so many interesting segments beyond those in newer,
- 1:23:35in the newer editions of AI.
- 1:23:37See what you did there.
- 1:23:38Even shouting out.
- 1:23:39ai.
- 1:23:39See what you did there.
- 1:23:40In conclusion, I want to reiterate that I can only give you five stars,
- 1:23:43but you truly deserve many more.
- 1:23:44I wanna take, I wanna take this opportunity to express my deep
- 1:23:47respect and admiration for the show's host Kevin and Kevin.
- 1:23:51Oh my God.
- 1:23:52Gavin and Kevin, their unique perspectives on ai, especially
- 1:23:56Gavin's a lot of value to the show.
- 1:23:58I eagerly anticipate hearing their insights on the fascinating
- 1:24:02news articles they share.
- 1:24:03Here's to Gavin, the president of the fan club I'm forming in my heart.
- 1:24:08that out.
- 1:24:08I should've sniffed that out.
- 1:24:09Gavin never wants to read the longer reviews, but he was
- 1:24:12adamant that he'd take this one.
- 1:24:13I get it.
- 1:24:14Okay.
- 1:24:14Hey, thank you Doc Anderson.
- 1:24:16I'm not AI, and Gavin is occasionally useful.
- 1:24:19That's fair.
- 1:24:20Alright, enjoy your victory dance.
- 1:24:21Thank you to everybody.
- 1:24:23Who took a second to engage.
- 1:24:24Whether you left us a review or whether you thought about it.
- 1:24:26Maybe you got that itchy scrolling finger.
- 1:24:28Just tap.
- 1:24:29Do it.
- 1:24:30Please.
- 1:24:30Subscribe.
- 1:24:30Follow.
- 1:24:31Like.
- 1:24:31Engage.
- 1:24:32Leave a comment.
- 1:24:33It's the only way we survive.
- 1:24:35Go play with Suno, everybody.
- 1:24:37Go have some fun this week , and try some different stuff out.
- 1:24:39, we have a great time doing this show and we will see you all next week.
- 1:24:42Another show, Kevin, in person coming next week.
- 1:24:46I've been practicing.
- 1:24:47I mean, I've been going to the Dave and Buster's and punching that
- 1:24:50speed bag arcade as hard as I can.
- 1:24:51I'm coming for you.
- 1:24:52I'm knocking that mustard cap right off your noggin, buddy.
- 1:24:55I can't wait.
- 1:24:56All right, everybody.
- 1:24:57Thanks.
- 1:24:57We'll see you next week.
- 1:24:58Bye bye